标签：#TL

Throttling: vì sao in-memory rate limit "biến mất" sau khi scale ngang, và chọn token bucket hay sliding window Throttling là cơ chế giới hạn số request một client (user, IP, API key, tenant) được xử lý trong một khoảng thời gian, để chống abuse, bảo vệ downstream, và phân bổ công bằng dung lượng service. Định nghĩa nghe đơn giản, nhưng lý do dev gặp nó trong việc thật lại rất cụ thể: sau khi scale service từ 1 pod lên 8 pod, cùng cấu hình "100 req/min mỗi user" đột nhiên trở thành 800 req/min thực tế — vì mỗi pod đếm riêng trong RAM, và load balancer rải request đều tám hướng. Rate limit vẫn "chạy", log không có lỗi, nhưng downstream vẫn bị flood. Đó là failure mode dẫn tới việc phải chuyển counter sang store phân tán, và kèm theo là câu hỏi chọn algorithm nào — token bucket, sliding window, hay leaky bucket — mỗi cái đánh đổi khác nhau. Cơ chế hoạt động Bốn thuật toán phổ biến, khác nhau ở cách đếm và cách xử lý burst. Fixed window counter. Chia thời gian thành khung cố định (mỗi phút bắt đầu tại giây 0). Mỗi request INCR một key rl:{user}:{minute} , nếu counter vượt limit thì reject. Đơn giản nhất, một INCR + EXPIRE trên Redis là xong. Nhược điểm cứng: tại biên khung có thể chịu gấp đôi limit trong một cửa sổ trượt — user gửi 100 req vào giây 59 của phút 12:00, rồi 100 req vào giây 01 của phút 12:01, tức 200 req trong 2 giây thật, trong khi limit là 100/phút. Sliding window log. Lưu timestamp của từng request trong sorted set, mỗi request ZADD + ZREMRANGEBYSCORE xoá các entry cũ hơn now - window , rồi ZCARD để đếm. Chính xác tuyệt đối nhưng tốn bộ nhớ tuyến tính theo số request. Sliding window counter. Cách Cloudflare mô tả trên engineering blog: giữ counter của khung hiện tại và khung trước, ước lượng lượng request trong cửa sổ trượt bằng nội suy có trọng số theo phần trăm khung trước còn nằm trong window. Chỉ tốn hai counter, sai số rất nhỏ so với log thuần, và không có failure mode biên như fixed window. Token bucket. Bucket có capacity B token, refill với tốc

Amazon says it has enough satellites in orbit to begin initial broadband service at mid-latitudes later this year.

"It works on my test input" is the most dangerous sentence in on-device AI development. I typed that sentence - or some version of it - a dozen times while building Redacto, our on-device PII redaction app running Gemma 4 E2B on a Samsung Galaxy S25 Ultra. The model would redact a patient name from a clinical note, I would nod, and I would move on. Then I would hand the phone to a teammate, they would type a police report, and the model would redact the suspect description instead of the victim name. The problem is not the model. The problem is that manual spot-checking is not validation. You are testing a single input against your own expectations, with all the confirmation bias that entails. When you have five domain modes (HIPAA, Financial, Tactical, Journalism, Field Service), three difficulty levels, and two candidate models, you need something systematic. You need a benchmark suite. This post covers how I built one - from dataset curation to scoring methodology to on-device infrastructure - for a hackathon app running entirely on a phone. No cloud. No API calls. No data leaving the device. Why Not Use an Existing Framework? The LLM evaluation space has mature tools. EleutherAI's lm-eval-harness is the community standard for evaluating language models against academic benchmarks like MMLU, HellaSwag, and ARC. Stanford's HELM (Holistic Evaluation of Language Models) provides a multi-metric evaluation framework with standardized scenarios. Google's BIG-bench offers hundreds of tasks for probing specific capabilities. These frameworks are excellent for what they do. They are also completely wrong for this problem, for three reasons. First, they assume server-side inference. lm-eval-harness expects to call a model through an API or load it in PyTorch on a GPU server. Redacto's model runs on a Qualcomm Hexagon NPU inside a phone. There is no Python runtime, no HuggingFace tokenizer at evaluation time, no way to hook into the framework's inference loop. Second, their

I fine-tuned Gemma 4 E2B. The adapter merged cleanly. The export to .litertlm completed without errors. I pushed the model to my phone, initialized the engine, and everything looked green. Then I tried to create a conversation and got this: Failed to apply template: unknown method: map has no method named get (in template:238) No model loading failure. No quantization error. The model initialized, the tokenizer loaded, and then the runtime choked on a Jinja template feature it does not support. This failure only surfaces when you actually try to run inference, not when you load the model. If you are demoing at a hackathon, this is the worst possible time to discover a compatibility issue. I hit this exact bug while building Redacto, a zero-trust PII redaction app that runs Gemma 4 E2B entirely on-device. This post walks through the full fine-tune-to-deploy pipeline: how to QLoRA a model on Colab, export it for LiteRT-LM, and avoid the undocumented template trap that will block your deployment. The Full Pipeline Here is what the fine-tune-to-deploy pipeline looks like end to end: HuggingFace base weights -> QLoRA fine-tune (Colab) -> Merge adapter into base -> Patch chat template <-- the step nobody tells you about -> Quantize + export to .litertlm -> Push to device Each stage has its own failure modes. The template patch step is the one that was undocumented at the time, and it is the one that will cost you hours if you do not know it exists. A note on framing before we dig in: this was an under-resourced fine-tune. I trained on 3,000 of the 400,000 samples in the ai4privacy/pii-masking-400k dataset for a single epoch, and the label format did not fully match what Redacto expected downstream. The point of this post is not the fine-tune's accuracy - it is the deployment mechanics I had to work through to get any fine-tuned model onto the device at all. Step 1: QLoRA Fine-Tuning on Colab QLoRA (Quantized Low-Rank Adaptation) lets you fine-tune a quantized model by tra

Here's a bug that looks impossible until you understand Gson. You have this data class in your Pokedex app: data class PokemonStat ( @SerializedName ( "base_stat" ) val baseStat : Int , val stat : StatInfo ) A teammate cleans up the code and deletes what looks like a redundant line: data class PokemonStat ( val baseStat : Int , // @SerializedName removed val stat : StatInfo ) It compiles . No red errors. He runs the app — and every Pokemon's baseStat is 0 . Not the real 45 or 49. Just 0 . Everywhere. And Gson never threw an error, never logged a warning, never said a single word. If you can explain why this happens, you understand Gson better than most juniors. This is also a favorite interview question. Let's break it fully. What Gson actually does When the JSON comes back from the server, Gson goes key by key: Read a key from the JSON — say base_stat . Look for a Kotlin property with the exact same name . Found it? Pour the value in. Not found? Leave that property at its default value and move on. That's the entire matching game — exact name match, or nothing. Now look at the "cleaned up" class. The JSON key is base_stat . The Kotlin property is baseStat . Those are not the same string . Gson looks for a property called base_stat , doesn't find one, shrugs, and leaves baseStat at its default. The default for an Int is 0 . That's your bug. @SerializedName("base_stat") was never redundant. It was the sticky note telling Gson: "this property is called baseStat in Kotlin, but look for base_stat in the JSON." Delete the note, and Gson stops matching. Two ways to fix it: Rename the property to base_stat — works, but breaks Kotlin's camelCase convention. Put @SerializedName("base_stat") back — keeps the clean name and matches. This is the right one. But why 0 ? Why not a crash? This is the part that surprised me. A missing field feels like it should be an error. It isn't. Gson treats a missing key as allowed . It builds your object, fills the fields it found, and leaves

In 2024, Apple shipped something genuinely great: AirPods Pro can run a clinical-style hearing test and then act as hearing assistance, tuned to your ears. People love it. There's just one catch — it needs an iPhone to set up, recent AirPods to run, and if you're on Android you get nothing. Meanwhile, the average pair of prescription hearing aids costs about $4,700 , and surveys show a $1,500 device is simply out of reach for more than half the people who need one. There are a billion-plus Android phones out there, most of them sitting next to a pair of ordinary earbuds that already contain everything you physically need: a microphone, a DAC, and speakers. The gap seemed absurd. So I've spent the past weeks building OpenHearing — a free, GPLv3 Android app that does the whole pipeline: Hearing check — a pure-tone test using the modified Hughson–Westlake staircase (the same adaptive up-down procedure audiologists use), per ear, per frequency. Or skip it and type in the numbers from a real audiogram. Sound profile — the results are fitted into a per-ear gain curve (half-gain rule for v1; NAL-NL2 is a pluggable strategy for later). Real-time assist — mic in, per-ear DSP, earbuds out. Quiet speech gets louder. Works with whatever earbuds you already own. No root, no special hardware. It is very deliberately not a medical device — no diagnosis, no treatment claims, big disclaimer before anything plays a tone. Think of it as the open, inspectable "gateway" tier below real hearing care. This post is about the three engineering decisions that turned out to matter most. 1. The safety-critical DSP is pure Kotlin — and that's the whole point An app that amplifies sound directly into human ears has exactly one unforgivable failure mode: being loud when it shouldn't be. So the entire signal chain is plain Kotlin with zero Android dependencies, hidden behind a tiny I/O interface. AudioRecord / AudioTrack is a dumb shell; everything that can hurt someone is JVM-testable: input → EQ

It happened during a quiet, solemn moment in a community prayer hall. I was sitting in the third row, reflecting, when suddenly, a high-pitched ringtone shattered the silence. It wasn't my phone, but the ripple effect of embarrassment was immediate. Everyone looked around, shifting uncomfortably. That collective tension is something we have all felt—a moment of human error that technology should have intercepted. I looked at my own device, feeling the familiar anxiety of whether I had remembered to flip the physical silent switch. It was then that I decided to stop relying on my own memory. We live in an era of hyper-connected devices, yet the most basic context-awareness—knowing where we are and how our phone should behave—remains manual. I found myself constantly toggling between 'Normal' and 'Silent' modes at the library, the office, and the gym. If I forgot, I was the person disrupting a meeting. If I remembered to mute it, I inevitably forgot to unmute it, missing urgent calls from family for hours. The existing solutions were either too bloated, requiring invasive cloud permissions, or they simply failed to trigger reliably when the screen was off. I needed a solution that was local, predictable, and battery-conscious. Building Muffle started with the realization that I had to master the GeofencingClient API without draining the user's battery. The primary challenge wasn't just triggering an event; it was doing so while the device was in a deep sleep state. I initially experimented with a standard LocationManager approach, polling GPS coordinates at set intervals. That was a disaster. It kept the radio active, pinged the GPS satellites constantly, and decimated the battery life in under four hours. It was an immediate non-starter for a production-ready application. I pivoted to the Geofencing API provided by Google Play Services, which leverages the fused location provider. This is significantly more efficient because the system handles the batching and hardwa

NASA awarded Rocket Lab deals for three dedicated launches using the company's Electron rocket.

A client wanted to know who was on their site and on which page, the way Tidio's widget showed them — but without paying for a Tidio subscription, on an external WordPress site that doesn't use Firebase. The result: an external tracker hooked to an independent Firebase project, live presence via onDisconnect , persistent history in Firestore with IP geolocation — and a final debugging session where a browser CORS error was masking a server crash caused by an empty string instead of null . The context The client already had a third-party live-chat script installed on their site, and that's where the idea came from: "can we see who's on the site and on which page, without using that service?" Two constraints made the request less trivial than usual: the site hadn't been migrated to my usual stack yet — it was still running on WordPress, on different hosting — and there was no intention of introducing Firebase on the WordPress side. Step 1 — understanding what a live-chat widget actually does Before building anything, it was worth looking at what the already-installed script actually did. The tag pasted into the site was just a small loader: it creates a hidden iframe, loads the widget's real "brain" inside it from the provider's servers, which then connects via websocket to their backend to stream presence, current page and events in real time. The interesting part — "see who's on the site and on which page" — isn't in the public script: it all lives server-side at the provider, behind authentication, a proprietary dashboard and a subscription. There was nothing to "detach" from that service: it's client code tied to someone else's backend by design. But the pattern itself — a script tag hooking into an external backend — is exactly what's needed to build the same feature independently, and it fits well with Firebase, which has a native presence mechanism built for precisely this. Step 2 — live presence with Firebase Realtime Database Firebase Realtime Database has a

If you're going to apply for Startup Battlefield Australia, now is the time. Applications close July 6, and once the deadline passes, the opportunity is gone.

The Federal Bureau of Investigation (FBI) said today it worked with industry partners to seize hundreds of domains associated with NetNut, a sprawling residential proxy service operated by the publicly-traded Israeli company Alarum Technologies [NASDAQ: ALAR]. The action comes roughly two weeks after KrebsOnSecurity published findings from multiple security firms connecting NetNut to the Popa botnet, a collection of at least two million devices that have been compromised by malicious software with little or no consent from victims.

Hopper will pay $35 million to settle FTC allegations that it used deceptive “dark patterns” to hide fees and mislead travelers about the cost and benefits of services.

Opening hook It happened during a quiet afternoon at the mosque. The imam was mid-sentence when a rhythmic, high-pitched ringtone cut through the silence like a knife. Every head turned. It was my phone. My heart sank as I scrambled to silence it, only to realize I had forgotten to flip the physical toggle before walking in. That moment of collective, disappointed glares burned. It wasn't just an annoyance; it was a total breakdown of my focus and a social failure I had accidentally caused because my phone couldn't manage itself. The problem We live in an era where our devices are supposedly 'smart,' yet they are remarkably bad at knowing when to keep quiet. We carry computers in our pockets that can calculate the exact position of the moon or stream 4K video, but they cannot inherently tell that we are in a meeting, a lecture, or a place of worship. You could argue that setting a manual schedule works, but life isn't static. Meetings run over, prayer times shift by a minute each day based on astronomical calculations, and spontaneous plans happen. I found myself constantly juggling the physical volume buttons. If I remembered to mute it, I inevitably forgot to unmute it afterward, missing urgent calls from family. If I didn't mute it, I was the person disrupting the room. I wanted a solution that respected the context of my location and the specific time of day without requiring me to touch my screen. The core friction is that Android is designed to restrict background processes to save battery, which is exactly what a silent-automation app needs to thrive. Getting the app to reliably trigger a volume change while the phone is sitting in a pocket, deep in Doze mode, became my primary development hurdle. The technical decision / implementation When I started building Muffle, I initially tried a standard Service with a Handler loop to check conditions. It worked fine while the screen was on, but as soon as the phone entered Doze mode, the OS aggressively throttled my

What if one pitch changed everything? The next company nobody has heard of yet is building something that will matter. It could be yours.

Force nginx to TLS 1.3 An nginx server is accepting an old TLS version, and the exam wants it locked to TLS one point three. The config lives in a ConfigMap. The catch is that editing the ConfigMap alone changes nothing. Let's do it the way the CKA expects. 🎥 Watch the video: https://www.youtube.com/watch?v=rx-77YBw99w This is a CKA Workloads & Scheduling walkthrough. Every command below is real output from a live cluster, and you can reproduce the whole thing yourself (scripts at the end). The scenario An nginx-static Deployment serves HTTPS, and its server config comes from a ConfigMap named nginx-config. Right now it allows both TLS one point two and one point three. Your task is to allow only TLS one point three, then make nginx actually use the change, so that a TLS one point two request fails. nginx-static serves HTTPS from the nginx-config ConfigMap It currently allows TLS 1.2 AND 1.3 Restrict ssl_protocols to TLS 1.3 only A TLS 1.2 request to the Service must then fail How nginx, ConfigMaps, and rolling restarts fit together Two ideas drive this. First, ssl_protocols is an allow list; leave only TLSv1.3 and nginx rejects any older handshake. Second, a ConfigMap mounted into a pod updates the file on disk, but nginx only reads ssl_protocols when it starts. So you must roll the Deployment, with kubectl rollout restart, for the new value to take effect. Inspect the current state Start by seeing what is running and what the config says. The nginx-static Deployment, its Service on port four forty three, and the nginx-config ConfigMap are all here. Grep the rendered ConfigMap for the ssl_protocols line: it lists TLSv1.2 and TLSv1.3, so old clients still get in. $ kubectl -n nginx-static get deploy,svc,configmap NAME READY UP-TO-DATE AVAILABLE AGE deployment.apps/nginx-static 1/1 1 1 17h deployment.apps/tester 1/1 1 1 17h NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE service/nginx-static ClusterIP 10.96.13.162 <none> 443/TCP 17h NAME DATA AGE configmap/kube-root-ca.

Your CI catches the npm vulnerability. Your developer is already three branches away and one standup behind. The package is installed, the lockfile regenerated, the import wired into a service, and the human who made that decision did it on a Tuesday afternoon with a tab open to Stack Overflow. Now the scanner is yelling. From the terminal, that is not security. That is grief counseling. That is the frame Sonu Kapoor lays out in a DevOps.com essay this week, and the engineering bones of it are correct. A scanner is not a gate. It is a status check. Kapoor's argument is about feedback loops. A developer installs, codes, commits, pushes. Only then does CI run. By the time the finding surfaces, the decision to add the package, and the context for why, has evaporated. So has the lockfile churn that caused it. What started as "is this package safe?" becomes "fix this in a different sprint." The scanner did its job. The fix is now a project. He backs it with a small case study from the NestJS repo: a scan of package-lock.json returned 1,626 resolved packages and 25 vulnerabilities. Of those, 12 were directly fixable. Thirteen were transitive, buried in upstream graphs, waiting on someone else's release. In a pipeline-first workflow, every dependency hop is a separate commit and a separate run. (Multiply by the number of services your team owns. Then by your runner-minutes budget. Send me the bill.) The arithmetic gets ugly quickly. A single lockfile with more than fifteen hundred resolved packages is not exotic for a working Node app, it is the default. The chance that the first time anyone looks at that graph is during a pipeline run, after the merge intent is already in the reviewer's queue, is the structural bug. Where the essay is right, and where it gets too tidy Concede the obvious. CI is not the problem. CI is fine. It runs uniformly, it cannot be skipped, and it is the right place to fail a build when an OSV record drops mid-week against a dependency that was clea

The Kotlin Multiplatform (KMP) ecosystem moves fast. To stay at the cutting edge, ImagePickerKMP has recently undergone a major architectural upgrade in version 1.0.42 , adopting the latest stable versions of Kotlin and Compose Multiplatform. For the latest requirements and installation guides, always refer to https://imagepickerkmp.dev/ . Major Version Upgrades The v1.0.42 release brings significant updates to the core dependencies of the library: Dependency New Version Previous Version Kotlin 2.3.20 2.1.x Compose Multiplatform 1.10.3 1.9.x Ktor 3.4.1 3.0.x Android Gradle Plugin 8.13.2 8.x Warning: Kotlin 2.3.x brings breaking ABI changes. Projects using Kotlin < 2.3.x will fail to compile with an "ABI version incompatible" error when using ImagePickerKMP 1.0.42. Why the Upgrade Matters Performance: Kotlin 2.3.20 includes numerous compiler optimizations that result in smaller and faster binaries for both Android and iOS. Stability: Compose Multiplatform 1.10.3 resolves several rendering issues on iOS and Desktop, providing a smoother user experience. Future-Proofing: By moving to these versions, ImagePickerKMP is ready for the upcoming features in the Kotlin roadmap. How to Upgrade Your Project To use the latest version of ImagePickerKMP, you must update your build.gradle.kts file: plugins { kotlin ( "multiplatform" ) version "2.3.20" id ( "org.jetbrains.compose" ) version "1.10.3" } dependencies { implementation ( "io.github.ismoy:imagepickerkmp:1.0.42" ) } If your project is not yet ready for Kotlin 2.3.x, you can continue using version 1.0.41 of the library, which maintains compatibility with older Kotlin versions. Conclusion Staying updated is crucial for security, performance, and developer happiness. ImagePickerKMP makes it easy to leverage the power of the latest Kotlin features while maintaining a simple, unified API for media picking. Explore the full API reference and new features at https://imagepickerkmp.dev/ . References [1]: ImagePickerKMP Documentati

nginx Event Loop — Complete Lifecycle Reference A precise, bottom-up reference covering every buffer, syscall, interrupt, and data movement from the moment a TCP packet hits the NIC to the moment a response is sent back. Two concurrent users are used throughout as a concrete example. Table of Contents Foundations — fd and Socket Hardware Layer — NIC, DMA, Interrupts Kernel Structures and All Buffers epoll — How the Worker Waits Efficiently nginx Startup Sequence Complete Request Lifecycle — Two Concurrent Users What Happens While Worker is Busy All Buffers — Master Reference All Syscalls — Master Reference Failure Modes 1. Foundations 1.1 Everything is a File Linux's core philosophy: every I/O resource — files on disk, network connections, pipes, terminals, devices — is represented as a file. This means one unified API ( read , write , close ) works on all of them. The kernel manages the actual resource. Your process holds a token. 1.2 File Descriptor (fd) A file descriptor is just an integer . It is a per-process token that refers to a kernel-managed resource. The kernel maintains a table per process called the fd table — a simple array where the index is the fd and the value is a pointer into the kernel. Process fd table: ┌─────┬───────────────────────────────┐ │ fd │ points to │ ├─────┼───────────────────────────────┤ │ 0 │ stdin │ │ 1 │ stdout │ │ 2 │ stderr │ │ 3 │ listen socket (nginx) │ │ 5 │ User A client connection │ │ 6 │ User B client connection │ │ 12 │ backend connection for User A │ │ 13 │ backend connection for User B │ └─────┴───────────────────────────────┘ 0, 1, 2 are always pre-assigned. Application fds start from 3 upward. The fd is meaningless on its own. It only means something when passed to a syscall — the kernel uses it to look up the real resource. 1.3 Socket A socket is the kernel's internal data structure representing one end of a network connection. Created when your process calls socket() . Lives entirely in kernel RAM. Your process nev

It was the middle of a Friday afternoon, and I was sitting in the front row of a local mosque. The room was deathly quiet, the kind of silence that amplifies every heartbeat. Suddenly, three rows behind me, a phone erupted with a loud, brassy ringtone that seemed to go on for an eternity. The man scrambled to silence it, his face turning bright red as he fumbled with his screen. I felt his humiliation deeply. In that moment, I realized that modern smartphones—despite their intelligence—are remarkably stupid when it comes to context-aware social etiquette. We live in a world of smart devices, yet we are still manually toggling our volume settings like it is 2005. I have spent years forgetting to silence my phone before a meeting, a lecture, or a quiet space, only to have it buzz loudly at the worst possible time. It is a friction point that feels trivial until it happens to you, at which point it becomes incredibly disruptive. Existing solutions often fall into two camps: over-engineered automation tools that require a computer science degree to configure, or basic calendar-sync apps that lack the nuance needed for things like location-based triggers or recurring religious observances. I wanted something that just worked, quietly, in the background, without requiring me to constantly open an app to double-check if my rules were still active. When I started building Muffle, I quickly realized that the greatest obstacle wasn't the logic of detecting a location or a prayer time—it was the operating system itself. Android, in its quest to squeeze every millisecond of battery life out of a device, has turned into a minefield for developers trying to keep background tasks alive. If you rely on a standard Service , the system will kill it within minutes as soon as the user turns the screen off. I needed a way to ensure that my background monitoring, especially for geofencing and prayer time calculations, stayed alive even when the phone was sitting in a pocket for hours. I

We have all been there: you build a utility app that relies on precise location or time-based triggers, only to find that it works perfectly on your Pixel but dies silently on a Samsung or Xiaomi device. When I started building Muffle, an app designed to automate sound profiles based on prayer times and GPS, I realized that standard AlarmManager usage wasn't enough to survive aggressive battery optimizations. The Problem with OEM Kill-Switches Modern Android versions enforce strict background execution limits. If your app isn't a high-priority foreground service, OEMs will frequently kill your process to save a few milliwatts of battery. For Muffle, if the process dies, the user misses their silent profile trigger, which defeats the entire purpose of the app. I had to move away from relying on a long-running background service and rethink my architecture entirely. Moving to WorkManager with Expedited Jobs Instead of a persistent service, I transitioned the core logic to WorkManager . By utilizing ExistingPeriodicWorkPolicy.UPDATE , I ensure that the scheduling remains consistent even across reboots. However, WorkManager alone can be delayed by Doze mode. To combat this, I implemented setExpedited(OutOfQuotaPolicy.RUN_AS_NON_EXPEDITED_WORK_REQUEST) for critical profile switches. This tells the system that the work is time-sensitive. kotlin val workRequest = PeriodicWorkRequestBuilder(15, TimeUnit.MINUTES) .setConstraints(Constraints.Builder().build()) .setExpedited(OutOfQuotaPolicy.RUN_AS_NON_EXPEDITED_WORK_REQUEST) .build() Leveraging Foreground Services with Notifications For features requiring immediate precision—like geofencing—I had to accept that a persistent notification is non-negotiable. To keep the app from being perceived as 'spammy,' I designed the notification to be low-priority, showing only when a profile is actively being managed. I also had to handle the onTaskRemoved callback in my Service implementation. By calling startService again with a sticky